The partial observability probit model is a statistical model for discrete outcomes caused by a complex combination of multiple latent factors. It is useful for political science research because political scientists often study interactions of unobservable decision making by several actors or survey responses resulted from a mixture of psychological factors, and because outcomes are recorded as a discrete variable in many cases of political science. I introduce this model as well as its underlying models, and its application and extension in political science literature. In addition, I developed an applied model with partial observability for the study of the survey responses on ideological self-identification. Ideological self-identification is measured by where a respondent place oneself on a discrete ideological scale, and can be decomposed into three latent factors: recognition, extremity, and direction. The new model can be estimated by Markov chain Monte Carlo methods. I applied my model to Japanese opinion poll data. An information criteria judged my model was superior to the previous ones, and I found some results that could not be led by the previous models.
In this paper, we introduced a method of causal inference for arbitrary treatment regimes. We first expounded the generalized propensity score (Imai & Van Dyk, 2004) that extended the propensity score for binary treatments, and then causal inference with this generalized propensity score. Furthermore, we reanalyzed the data of relations between mass media and politics from a previous study, and discussed the usefulness of causal inference using generalized propensity scores.
The same human behavior arises for different reasons, and different behaviors have the same motivation. Behavioral data alone is not sufficient to identify psychological mechanisms underlying human behavior. The fMRI experiments provide researchers with opportunities to differentiate the important psychological processes. In this paper, we introduce our recent neuroimaging study on repeated prisoner's dilemma games. We have specifically explored the neural correlate of reciprocal interaction that has an immediate relevance to investigate the real dynamics in the society. Based on the evidence from the experiment, we examine the contribution of the neurocognitive approaches to deepening our understanding on human behavior. The article suggests the validity of the method and empirical knowledge in political science when we conduct the experiments in neuroscience.
Spatial analysis and modeling have played significant roles in the empirical analysis for political science. This paper first reviews the development and application of the spatial analysis in political cognition from methodological point of view. The emphasis is laid on the multidimensional scaling technique and its extensions. Then, one of the recent developments in special analysis, Bayesian K-INDSCAL, which is a multidimensional scaling with both individual and group differences, is applied to the Japanese expert survey data on party policies. Three latent-classes have been identified, with related but heterogeneous spatial representations. Individual differences in terms of dimensional weight have also been identified. Implications and future directions of spatial analysis in political science are discussed.
We propose an econometric two-stage model for category-level purchase and brand-level purchase which allows simultaneous brand purchase at the same time. The proposed model formulation is consistent with the traditional theory of consumer behavior, and the utility functions remain to be normally distributed. Such modeling approaches have not been found in existing econometric models. The simulation studies show the previously proposed related models can cause severe bias in predicting the future brand choices, while the proposed method can effectively predict them. Additionally in real data analysis, while the existing methods provided the parameter estimation results that were implausible, the proposed method provided the results that were plausible.
This study proposes a graded response model including a new parameter, which represents rater effects affecting inter-item variance. The validity of this model was shown by parameter restoration simulation, in which artificial item response data were generated and calibrated under 12 conditions each. Another simulation study, which compares estimation accuracy of θ in the presented model and original graded response model, was conducted. Results indicated the proposed model can remove the rater effect. Using a rating experiment that induced the halo effect, it was shown that the new parameter wir can be regarded as a halo effect.
We proposed a new method for analyzing the relationship among industrial sectors by Z-score standardizing and visualizing the input-output tables as minimum spanning trees (MST). These results are also quantitatively evaluated in terms of degree centrality and closeness centrality. We confirmed the effectiveness of our method in comparison with the simple method of no Z-score standardization and with a conventional method based on stock market data. In our experiments, using the input-output tables in Japan in 2000 and 2005 and stock data in 2005, the simple method centralized industrial sectors, (such as other business services whose trade volume is simply larger), while our proposed method centralized sectors, (such as electronics, financial and pharmaceutical industries). In addition, we confirmed that our method showed more accurate economic transaction between industries than the conventional stock analysis method.